模型发布/更新
-
Google DeepMind releases Gemini Robotics 2 physical AI
One brain. For any robot. 🤖 We are launching Gemini Robotics 2: our next-generation physical AI, bringing whole-body…
-
How GPT-5.6 advances the price-performance frontier
OpenAI has introduced lower pricing for the Luna and Terra versions of GPT-5.6, helping enterprises deploy AI…
-
Gemini Robotics ER 2: empowering robots with video understanding, task orchestration and multi-robot collaboration
Google DeepMind has introduced Gemini Robotics ER 2, a Gemini-based robotics foundation model. The model achieves…
产品发布/更新
-
Token Saver: open-source MCP extension cuts Claude PDF token consumption by 92%-99% with local hybrid RAG
The Marktechpost AI team released Token Saver, an open-source MCP extension for Claude Desktop that retrieves PDFs…
-
Gemini Spark integrates Chrome auto-browse feature
Gemini Spark 🤝 @GoogleChrome Gemini Spark is now integrated with Google Chrome's auto-browse feature. With your…
-
Google Earth integrates Nano Banana 2 image generation
Google Earth for web launches an image generation feature based on Nano Banana 2, allowing users to combine satellite…
-
Perplexity Computer launches Projects feature
Introducing Projects on Perplexity Computer. With the launch of Projects, we are transforming Computer into a…
-
GitHub Copilot app adds stacked sessions and pull request features
The GitHub Copilot app introduces stacked sessions, allowing users to create a series of sequential tasks in the same…
-
LangSmith 推出 Align Evals:校准 LLM 评估器以匹配人类偏好
LangSmith 发布新功能 Align Evals,帮助用户校准 LLM 评估器,使其更贴合人类偏好。该功能旨在简化评估流程,减少人工标注与自动评估之间的偏差,提升评估结果的可信度。
-
LangSmith LLM Gateway:将运行时治理内置于智能体生命周期
LangSmith 推出 LLM Gateway,将支出限制、PII 脱敏和追踪连续性等运行时治理能力直接内置于 LangSmith 平台。该网关专为 AI 智能体生命周期设计,可在不中断追踪的前提下对模型调用实施实时管控。
行业动态
-
Judge says Trump administration still lacks evidence to label Anthropic a supply chain risk
US District Judge Rita Lin said the Trump administration failed to provide sufficient evidence to justify labeling…
-
FCC bans imports of new Chinese robots and connected inverters
The U.S. FCC has banned imports of new Chinese "advanced robotic devices" and connected power inverters starting July…
-
Anthropic discloses Claude breached real systems during safety evaluation
Anthropic's cybersecurity evaluation review found that in three separate incidents, Claude models connected to the…
-
RadixArk partners with Google Cloud to bring full SGLang capabilities to TPU
RadixArk has partnered with Google Cloud to bring the open-source inference framework SGLang to Google TPU. Developers…
论文研究
-
Tencent Hunyuan's Hyra solves 50-year-old math problem
Tencent Hunyuan, using research agent Hyra and the Hy3 model, constructed an integer set A such that the exponent…
-
降维遇见网络科学:UMAP的kNN图在数据理解中的应用
Apple研究团队展示了UMAP内部kNN图在数据理解中的潜力,该图在原始高维空间编码数据流形。将PageRank、k-core分解和聚类系数等图算法应用于该图,可识别代表性数据点、揭示密集核心与稀疏边缘。在MNIST和Fashion M…
-
MoMo:通过时空动作分词实现机器人操作中的运动模式控制
机器人操作需根据任务、物体和交互环境调整动作执行方式。MoMo 提出两阶段模仿学习框架,包含时空动作分词器和行为克隆 Transformer,将任务与连续运动模式条件作为输入。在六项真实机器人操作任务中,改变该条件可稳定生成不同执行风格的…
技巧与观点
-
OpenAI President Brockman admits new ChatGPT desktop app is "a bit messy," targets "zero tabs" by year-end
OpenAI co-founder and President Greg Brockman admitted that the new ChatGPT desktop app after merging Codex has a "a…
-
如何使用 Google TPU 微基准测试评估 TPU 性能
Google 开源 TPU 微基准测试套件提供网络、计算、HBM、主机传输和注意力组件的细粒度性能指标,帮助开发者验证真实硬件能力。通过基准测试建立 Roofline 模型,工程师可准确诊断机器学习工作负载是受计算、内存还是网络限制。该经…
-
Cursor 如何为云智能体构建开发环境
Cursor 将开发环境本身作为面向智能体的产品来构建,使云智能体能测试代码变更。团队通过统一跨平台工具链、开发 CLI 工具 anydev 简化构建命令,并构建 Cursor Cloud MCP 实现环境自愈。目前,云智能体在 Curs…
-
cdnjs 迁移至 Cloudflare 开发者平台
2026 年 6 月 23 日起,开源 CDN 服务 cdnjs 完全运行在 Cloudflare 开发者平台上。该平台日均处理 108,000 次请求/秒、90 亿次请求,缓存命中率 98.6%。LLM 如 ChatGPT 和 Clau…
-
Databricks:构建AI优先医疗组织的基础
Databricks提出医疗组织推进AI计划时需应对“噪音”挑战,强调以数据基础设施为核心构建AI优先架构。该框架聚焦于整合分散的医疗数据、建立统一治理层,并支持临床与运营场景的模型部署。具体方案包括利用Lakehouse架构实现实时数据…
-
Skyscanner 如何用 Runway 消除前期制作中的猜测
Skyscanner 品牌团队在美加品牌广告拍摄中,使用 Runway 辅助艺术指导,在开拍前锁定镜头构图、灯光方向和演员站位,并在片场实时验证拍摄内容。团队将 Runway 生成的场景、道具和构图直接放入最终广告版式(OOH、Meta…